Abstract
Mangrove forests provide essential ecosystem services, including shoreline protection, carbon sequestration, and habitat for diverse coastal communities, but they are increasingly threatened by climate change, altered hydrology, and rising salinity. Traditional field-based sampling is labor-intensive and spatially limited, while satellite remote sensing often lacks the spatial and spectral resolution needed to detect early physiological stress before visible decline occurs. This study evaluated the potential of leaf-level spectral signatures with data transformations and vegetation indices to differentiate mangrove species and identify physiological stress associated with changing environmental conditions. Leaf spectra were collected monthly from March through December 2025 for Rhizophora mangle, Avicennia germinans, and Laguncularia racemosa across different locations in Estero Bay, Florida. Simultaneous measurements of soil salinity, pH, soil temperature, and leaf chlorophyll (SPAD) were used to examine relationships between environmental conditions and spectral responses. Raw, first derivative, and continuum-removed spectra were analyzed using PCA, PLS-DA, and RF models. To eliminate geographic data leakage and micro-canopy noise, a rigorous framework utilizing tree-level monthly means and a strict Leave-One-Location-Out (LOLO) spatial cross-validation scheme was implemented. The results demonstrate that when leaf-level pseudoreplication is suppressed, standard linear models perform exceptionally well; the spatially validated PLS-DA configurations achieved true validation accuracies between 99.62% (continuum-removed) and 100.00% (raw reflectance and first derivative) across all three species simultaneously. Non-linear Random Forest architectures similarly demonstrated robust classification performance (99.52% mean accuracy) and confirmed high cross-seasonal transferability (99.44%"--" 100.00%), demonstrating that species-specific spectral signatures remain temporally stable across varying environmental gradientsFurthermore, select spectral bands within the shortwave infrared and visible blue regions of the electromagnetic spectrum showed higher sensitivity for species discrimination. Results also highlighted species-specific salinity thresholds. These findings demonstrate that leaf-level hyperspectral spectroscopy can detect species identity and early physiological stress before visible symptoms of decline occur. By identifying diagnostic wavebands linked to both species composition and stress responses, this study provides a foundation for scaling hyperspectral monitoring from field measurements to airborne and satellite platforms, supporting proactive monitoring and conservation of vulnerable mangrove ecosystems.
